Aggregating Phillips Curves *

The New Keynesian Phillips Curve is at the center of two raging empirical debates. First, how can purely forward looking pricing account for the observed persistence in aggregate inflation. Second, price-setting responds to movements in marginal costs, which should therefore be the driving force to observed inflation dynamics. This is not always the case in typical estimations. In this paper, we show how heterogeneity in pricing behavior is relevant to both questions. We detail the conditions under which imposing homogeneity results in overestimating a backward-looking component in (aggregate) inflation, and, potentially underestimating the importance of (aggregate) marginal costs for (aggregate) inflation. We provide intuition for the direction of these biases, and verify them in sectoral French data. Our analytics identify the sources of heterogeneity most likely to result in aggregation bias under the two standard (homogeneous) estimators used to test the New Keynesian Phillips curve, the Generalized Method of Moments and the Maximum Likelihood approach. Our econometrics results provide a simple blueprint which, if disaggregated data are available, makes it possible to identify and correct for heterogeneity biases, and to predict their magnitudes depending on the source and extent of industry-level specificities. JEL Classifications: C10, C22, E31, E52.